mcp-realtime-voice

MCP Realtime Voice

Turn Claude into a voice assistant that listens and speaks. This MCP server handles speech recognition and text-to-speech for a natural voice interface.

⚠️ IMPORTANT:

• This has only been tested on macOS.
• Launch Claude from the terminal with the command below instead of clicking the app icon. This fixes microphone permission issues.

Features

• Speech recognition with silence detection
• Text-to-speech for AI responses
• Voice Activity Detection using Silero
• Works on Windows, macOS, and Linux
• Audio device management
• Simple voice conversation interface

Prerequisites

• Python 3.8+
• A microphone and speakers
• MCP client (Claude)

Installation

1. Clone this repo:

   git clone https://github.com/yourusername/mcp-realtime-voice.git
   cd mcp-realtime-voice
   

2. Set up a virtual environment:

   python -m venv venv
   
   # On Windows
   venv\Scripts\activate
   
   # On macOS/Linux
   source venv/bin/activate
   

3. Install dependencies:

   pip install -r requirements.txt
   

   System dependencies:

   • Ubuntu/Debian: sudo apt-get install portaudio19-dev
   • macOS: brew install portaudio

Usage

Connecting to Claude

1. Launch Claude from the terminal:

   # On macOS
   /Applications/Claude.app/Contents/MacOS/Claude
   
   # On Windows
   # Use the path to your Claude executable
   start "" "C:\Path\to\Claude.exe"
   

2. Install the MCP server:

   mcp install voice_server.py --name "Realtime Voice"
   

   Or test with the MCP Inspector:

   mcp dev voice_server.py
   

Available Tools

• list_audio_devices: Shows all audio input/output devices
• listen_for_speech: Records and transcribes speech
• speak_text: Converts text to spoken audio
• voice_mode: Starts interactive voice conversation

Voice Conversation Mode

To start:

1. Connect the MCP server to Claude
2. Ask Claude to "enter voice mode"
3. Start talking - the system will:
   • Listen for your speech
   • Detect when you finish speaking
   • Send transcribed text to Claude
   • Speak Claude's response

To exit, just say "exit voice mode" or "stop voice mode".

Configuration

Edit these values in voice_server.py if needed:

• VAD_THRESHOLD: Voice detection sensitivity (default: 0.2)
• SILENCE_DURATION: Seconds of silence before recording stops (default: 3)
• Audio sample rate and format settings